News / Korea / yonhap_finance
Anthropic Exposes Chinese AI Firms' Attempt to Illicitly Clone 'Claude' Model
Technology Services · Internet Software/Services · yonhap_finance · 2026-09-10
Anthropic has revealed that Chinese AI companies, including Moonshot AI and DeepSeek, used illicit proxy services to scrape and distill its Claude model.
What Happened
Illicit Model Distillation: Anthropic has released a threat report detailing how Chinese AI firms, such as Moonshot AI and DeepSeek, attempted to clone its 'Claude' model. These companies utilized illegal 'transfer stations' and fake accounts to route millions of user queries through Claude to distill its responses for their own model training.
Industrial-Scale Data Theft: The report highlights that Moonshot AI used thousands of fraudulent accounts to interact with Claude over 23 million times between May and July. U.S. officials have previously characterized such distillation practices as industrial-scale theft, warning that companies involved could face sanctions or blacklisting.
Real-World Security Risks: Beyond model cloning, Anthropic identified instances where its AI was misused for dangerous activities, including attempts to modify avian influenza viruses and the creation of cyber-attack tools. These findings indicate that concerns regarding AI misuse have shifted from theoretical risks to tangible security threats.
Global Cyber Threats: The report also documented various global malicious activities, including state-sponsored hacking operations by Russian groups against Ukraine and the use of AI for election manipulation. Anthropic emphasized that these incidents underscore the urgent need for stricter oversight of AI model accessibility and usage.